Registration of Multi-Spectral Manuscript Images

Abstract

Two medieval Slavonic manuscripts are recorded, investigated and analyzed by philologists in collaboration with computer scientists. The aim of the project is to develop algorithms that support the philologists by automatically deriving the description and restoration of the scripts. The parchment partially contains two scripts, where the first script was erased. In order to enhance the erased script, the manuscript pages are imaged in seven bands between 330 and 1000 nm. A registration, aligning the resultant images, is necessary so that further image processing algorithms can combine the information gained by the different spectral bands. Therefore, the images are coarsely aligned using rotationally invariant features and an affine transformation. Afterwards, the similarity of the different images is computed by means of the normalized cross correlation. Finally, the images are accurately mapped to each other by the local weighted mean transformation. The algorithms used for the registration and preliminary results are presented in this paper.